Latin hypercube analysis of parameter sensitivity in a large model of outdoor recreation demand

Abstract A model of recreation demand in Nevada and Utah for the years 1982 to 1994 allocated visitor pool from each population center to each recreation site in the two states as a function of travel time, recreation preferences, and population. The model was very large with several matrices of nearly 100 000 elements and with 15 adjustable parameters. Parameter sensitivity and error analysis was performed using Latin hypercube sampling of parameter values and stepwise regression of five response variables on the parameter values. Latin hypercube sampling allowed the response variables to be generated in less than 1 2 h of computer time, and the stepwise regression provided valuable insight into the behavior of the model with respect to the parameters.